At long last and after several glitches L2 Soil moisture reprocessing is over!
As you may be aware, ESA reprocessed level 1C with a fixed calibration (finished in January 2011) but could not reprocess level 2 in a timely fashion (typically November 2011) as it would have required on their system – assuming it was operational- anything between 4 and 6 months, i.e., too late wrt the general reprocessing. Hence the idea to wait for the official reprocessing in LTA facilities once they are operational, which implies a end of 2011 start.
So it was decided with ESA that the reprocessing would be done by CESBIO, using the reprocessed level 1C and the V4 of le Level 2 soil moisture algorithm configured exactly as it would be at DPGS (to ensure same format and consistent data set) and this on the CNES computing facilities. Ocean group was not interested in reprocessing just now incidentally, so it concerns only soil moisture.
François was fully in charge of this mammoth task!
Here is were we are today:
1) all data from DPGS L1C has been ingested.
* L1C data has been tested over DOME C and corresponds extremely well to ground measurements (see post on January 30 2011)
* the L2 SM V4 processor worked like a charm and all data has been processed as of yesterday evening
2) A few numbers…
We have processed
* 9377 L2 products (including 7394 Fully polarised) for the period ranging from 12/01/2010 to 26/12/2010.
* which corresponds to 714345863 nodes with roughly 77% leading to a retrieval
* this used up 38858cpu hours (1600 days!!), and 46TB of memory(on average 4h and 5Go per product).
* input data takes up 3.2TB and outputs 1.4TB
3) We started the processing on 07/01/2011, (47 days overall) with 3 stops
- from 20/01 to 25/01 (5d) test in the CNES Computing building
- from 02/02 to 04/02 (3d) link to mass storage access issue (space where all the data was stored)
- from 09/02 to 15/02 (7d) as we were still waiting for the DPGS reprocessed L1c data finished on 18/01 at ESAC, which also disrupted the parallel processing
So the whole task took 31 days and,should the last batch from DPGS had been delivered in January (so as to have the 6 processes running at the same time but only shifted slightly), we can reasonably say that it would have taken 25 days (François’ first estimate!).
Now while François is resting a little (well .. doing something else), we are now intensively checking the outputs
In parallel Philippe used the Prototype to run tests on smaller data sets (USA, Africa and Australia) with a large number of algorithm versions and / or configurations (309, 410 , dual in full vs full , with and without currents etc) we are comparing all the stuff (and it takes time and people!)
conclusions -very preliminary we still have lots to do, are very encouraging and we can safely state now that we realised that we had made a couple of blunders at the beginning (to much haste!) which means that the results are probably slightly less good than they could be, but up to now (we still have a fair bit to analyse) results look fine. A bit more « noisy » maybe, especially Tau but closer to field measurements and more realistic.
To make a long story short improvements!
We are now very busy checking everything and hope to be able to show many things at the quality working group getting the green light for dissemination to you all through ESA’s usual accesses of this ESA/CNES CESBIO reprocessed levels 1C and 2 SM.
More on this next week with results … stay tuned.